MLBenchy is running few neural networks on the platform.
The application is pretty large, around 360MB because it includes 7 of the most commonly used inferences. We recommend that you only download it over wifi.
It is using very common machine learning algorithms and evaluate the level of performance of your device. Here are the details of the performance test:
Nudity detection model is 224x224, net size is 24MB
TinyYolo is 416x416, net size is 64MB
Restnet is 224x224, net size is 100MB
Sentiment is a Dictionary based inference, with a string to Double association , Net size is 275KB.
Car Recognition is 224x224 , net size is 25MB
GenderNet is 227x227, net size is 45MB
InceptionV3 is 299x299, Net size is 94MB
GoogleNetPlaces is 224x224, and net size is 25MB
1st iteration of the benchmark is loading the neuronal network into memory, while running the inference once. It shows you the warm up time of the CoreML system.
2nd iteration is having too much variation to be meaning full.
3rd iteration shows you a pretty stable number.
Please contact us if you need interpretation assistance.
This application will help you to make a purchasing decision, it will allow you to compare the new platforms with the previous one, and decide if you want to spend the extra money and upgrade.
- The Purpose of the app is to help customers/consumers and programmers to understand the level of performance of the
- The target audience is consumers interested to understand the level of performance of the platform and my friends in the online tech press interested to evaluate the future usage models enabled by the progress in machine learning across the platforms.
- The app is intended for the general public, to allow them to compare platforms and make an inform decision about the machine learning and architecture capabilities of the platform.
- The Application does not requires accounts, it is free to use for anybody without logging.
- One of the targeted audience is the online press, they did express the need to have this application available for the general public, as they require to get their own customers to verify the analysis they do about the computing platforms they are writing about.